CSE 450 - Machine Learning

Repository for Team NorthWind’s coursework in CSE 450 Machine Learning at BYU-Idaho.

Team Members

_{Caleb Dilley}

_{Dallin Wagner}

_{Jonathan Oliphant}

_{Nels Buhrley}

Module 2 - Bank Marketing Prediction

Objective: Predict which bank clients will subscribe to a term deposit, turning an unprofitable phone campaign into a profitable one.

We built three classifiers on the UCI Bank Marketing dataset (37k records, 11.4% positive rate). The core challenge is class imbalance – a model that always says “no” scores 89% accuracy but generates zero revenue. We tackled this with SMOTE oversampling, class weighting, and probability threshold tuning, evaluating models on business value rather than accuracy.

Model	Technique	Key Idea
RF + SMOTE	Random Forest	Synthetic oversampling + manual class weights
RF Balanced	Random Forest	Automatic balanced class weights
Stacking (RF + KNN)	Ensemble stacking	RF and KNN base learners feed a logistic regression meta-learner; threshold tuned to 0.61

Highlights

Campaign Value

Without ML filtering: the campaign loses money ($-157 on 410 test contacts)
With our best model: $824 profit on the same 410 contacts
Projected at scale (4,119 contacts): up to $7,775 in campaign value

Golden List / Black List

The models concentrate the call list on high-conversion groups (previously converted clients, students, retirees) and filter out low-yield contacts (landline-reached, blue-collar workers), boosting precision from 11.5% to 47.2%.

Full technical writeup, per-model breakdowns, and detailed results in module_2-bank/README.md.

Repository Structure

module_2-bank/       Bank marketing prediction project 
tools/               Shared utility scripts
notebooks/           Exploratory Jupyter notebooks
nels_b/              Nels's working directory
dallin_w/            Dallin's working directory
caleb_d/             Caleb's working directory
jonathan_o/          Jonathan's working directory